Radar echo extrapolation, a critical spatiotemporal sequence forecasting task, requires precise modeling of motion trajectories and intensity evolution from sequential radar reflectivity inputs. Contemporary deep learning implementations face two operational limitations: progressive attenuation of predicted echo intensities during autoregressive inference and spectral leakage-induced diffusion at high-intensity echo boundaries. This study presents RaDiT, a hybrid architecture combining differential transformer with adversarial training for radar echo extrapolation. The framework employs a U-Net backbone augmented with vision transformer blocks, utilizing differential attention mechanisms to govern spatiotemporal interactions. Our differential attention mechanism enhances noise suppression under high-threshold conditions, effectively minimizing spurious feature generation while improving metric reliability. A conditional GAN discriminator is integrated to maintain microphysical consistency in generated sequences, simultaneously addressing spectral blurring and intensity dissipation. Comprehensive evaluations demonstrate RaDiT’s superior performance in preserving spatiotemporal coherence and intensity across 0–90 min forecasting horizons. The proposed architecture achieves CSI improvements of 10.23% and 2.88% at 4 × 4 and 16 × 16 spatial pooling scales, respectively, for ≥30 dBZ thresholds on the CMARC dataset compared to PreDiff. To our knowledge, this represents the first successful implementation of differential transformers for radar echo extrapolation.

This study proposes a novel quality control method combining fuzzy logic and threshold discrimination for processing X-band dual-polarization radar data from Beijing. The method effectively eliminates non-precipitation echoes, including electromagnetic interference, clear-air echoes, and ground clutter through five key steps: (1) Identifying electromagnetic interference using continuity of reflectivity across adjacent elevation angles, radial mean correlation coefficient, and differential reflectivity; (2) Preserving precipitation data in ground clutter-mixed regions by jointly utilizing the difference in reflectivity before and after clutter suppression by the signal processor, and characteristic value proportions; (3) Developing a fuzzy logic algorithm with six parameters (e.g., reflectivity texture, depolarization ratio) for ground clutter and clear-air echoes removal; (4) Filtering echoes with missing dual-polarization variables using cross-elevation mean reflectivity, mean correlation coefficient, and valid range bin proportion; (5) Removing residual noise via radial/azimuthal reflectivity continuity analysis. Validation with 635 PPI scans demonstrates high identification accuracy across echo types: 93.5% for electromagnetic interference, 98.4% for ground clutter, 97.7% for clear-air echoes, and 98.2% for precipitation echoes.

Three ground-based radars in the Pearl River Delta successfully observed Typhoon Higos (2020), which traveled over the offshore area in the South China Sea. During the observation period, the stratiform region of the outer rainband of HIGOS became active while swirling inward, merging into an unclosed eyewall and spreading outward, but its structure was asymmetric between upwind and downwind. To understand the dynamic mechanism of the asymmetry of the stratiform region in detail, the refined wind speed distributions in the inner core of Higos was retrieved by using the radar observation data and a three-dimensional, variational, direct, data assimilation, Dual-Doppler analysis (DDA). In addition, an Observing System Simulation Experiment (OSSE) was conducted with the numerical simulations by the Weather Research and Forecasting (WRF) model and numerical emulations by Cloud Resolving Model Radar SIMulator (CR-SIM) software to validate the retrieved data. From the OSSE, the emulated retrieved data were comparable with the WRF-out data. The analysis shows that the dynamic mechanisms are different between upwind and downwind in the stratiform rainband. In the former, the inflow sinks in the middle troposphere. In addition, there is an inflow in the lower troposphere, with an outflow aloft the inflow. In the latter, however, the stratiform rainband is primarily influenced by outflow from inside the rainband and inflow from outside the monsoon-related southwesterly winds. The vertical velocity characteristics in the stratiform rainband downwind also differ from those upwind. The upwind updraft was distinct in the middle troposphere, whereas the downwind updraft was caused by the convergence of the outflow from inside the stratiform rainband and the monsoon-related southwesterly inflow in the lower troposphere.